Abstract:
Background: Intensive care unit (ICU) mortality remains a major challenge in low- and middle-income countries. Limited financial, human and infrastructural resources further constrain critical care delivery. Severity scoring and prognostic models such as the Acute Physiology and Chronic Health Evaluation II (APACHE II), Mortality Probability Model III at admission (MPM0 III), and Rwanda Mortality Prediction Model (R-MPM) can used to estimate ICU mortality. However, their uptake and utilization in African settings remains low.
Methods: We conducted a cross-sectional study among 416 adult patients admitted to the ICU at Moi Teaching and Referral Hospital (MTRH), Kenya, between October 2022 and July 2023. Data on demographic, clinical, physiological, and laboratory characteristics were collected within 24 hours of ICU admission. ICU mortality was the primary outcome. Model performance was assessed using discrimination by area under the receiver operating characteristic curve (AUROC) and calibration using the Hosmer–Lemeshow goodness-of-fit test.
Findings: The mean age of the study participants was 42 years (SD 20), and 57% of patients were male The ICU mortality rate was 29% (120/416). APACHE II demonstrated the best discrimination for ICU mortality (AUROC 0·77, 95% CI 0·72–0·82), followed by MPM0 III (0·72, 0·67–0·78), while R-MPM had moderate discrimination (0·62, 0·56–0·68). All three models demonstrated poor calibration (Hosmer–Lemeshow p<0·05). Higher mortality was associated with admission from medical wards, hypotension, hypoxaemia, metabolic acidosis, acute kidney injury, and need for mechanical ventilation.
Conclusion: APACHE II and MPM0 III showed acceptable performance in predicting ICU mortality in this Kenyan ICU as compared to R-MPM. However, all models require additional local calibration to improve their utility in triage and patient care.
